The Hardest Problem May Be Heat

Part 11 of 30 · Series date: September 8, 2026

Space can offer a cold background while leaving a busy computer dangerously hot.

A hot frying pan cools differently on a kitchen counter than it would in a vacuum chamber. On Earth, air carries heat away. In space, a computer cannot simply blow its waste heat into the surrounding atmosphere.

This is the first fact to remember whenever a proposal promises “free cooling in space.” Space offers a cold radiative background in suitable directions. Reaching that background with enough heat is an engineering task.

Follow the heat, not the slogan

Heat moves through solid materials by conduction, through moving fluids by convection, and through electromagnetic emission by radiation. A spacecraft can use internal fluid loops, heat pipes, and conductive structures. But the final rejection to its surroundings is principally radiative.

NASA's spacecraft thermal-control reference describes this balance, including incoming sunlight and Earth-related heating as well as outgoing radiation. NASA: thermal control.

For a computing platform, the path might run from chip to cold plate, through a heat-transport loop, and out through a radiator. A restriction anywhere along that path limits sustained performance.

A square-meter thought experiment

The Stefan-Boltzmann relation gives an idealized starting point: emitted thermal power per unit area is emissivity multiplied by a physical constant and absolute temperature to the fourth power.

For an illustrative surface at 300 kelvin (about 27°C), with emissivity 0.9 and an unobstructed view of a very cold background, emission is about 413 watts per square meter. At 350 kelvin (about 77°C), it is about 766 watts per square meter. These are calculated teaching examples, not design ratings for a real radiator. Incoming radiation and practical losses reduce net rejection.

At the lower temperature, rejecting one megawatt would require roughly 2,420 square meters of ideal emitting area. Whether both sides can radiate effectively depends on geometry; emitting area is not automatically the same as one panel's footprint.

The fourth-power dependence makes higher temperatures attractive for radiation. It does not mean engineers can safely turn up chip temperatures without limit.

The radiator is not the processor

A chip's maximum allowed temperature, the coolant temperature, and the radiator temperature are different quantities. Heat needs a temperature difference to move through real components. Poor contact, long transport paths, or insufficient flow can leave a chip too hot even when part of the radiator is relatively cool.

This is why quoting a processor temperature in a radiator equation can mislead. The relevant surface temperature must come from a workable thermal design.

Mass also matters. A large radiator needs structure, deployment mechanisms, protection against leaks or punctures, and a configuration that avoids unwanted heating. These supporting requirements belong in any launch-cost calculation.

A cloud can schedule around heat

Imagine an orbital platform approaching an unfavorable thermal condition. Its scheduler might reduce processor load, postpone a batch job, or move work elsewhere. This is a plausible operating strategy, not a substitute for adequate cooling hardware.

The economic effect depends on the customer promise. A flexible analysis job may tolerate a pause. A service sold as continuously available may need additional capacity to hide it.

Thermal protection therefore connects directly to revenue. Peak computing performance describes what hardware can do briefly. Sustainable useful throughput describes what the business can sell repeatedly.

Failure should not spread unchecked

A proposed modular system could isolate damaged thermal loops so one leak does not disable the entire platform. Extra pumps or passive paths might protect essential functions. Each measure adds weight or complexity, so it must be justified against the loss it prevents.

These are design tradeoffs rather than universal prescriptions. A short demonstration and a long-lived commercial platform may reasonably choose different levels of protection.

Why a radiator cannot hide behind the solar array

Solar arrays want a favorable view of the Sun. Radiators often benefit from avoiding direct sunlight and looking toward a colder environment. That creates a layout problem before the first chip starts working.

Imagine a platform unfolding two broad wings. One collects energy; the other rejects heat. If their arrangement lets the collector shade the radiator from sunlight, that may help. If the same arrangement blocks the radiator's view or exposes it to warm nearby surfaces, it may reduce heat rejection.

An attractive folded package must therefore become a workable three-dimensional machine. Its geometry must support pointing, deployment, thermal balance, and communications together.

This is one reason component improvements cannot simply be added together. A wonderfully light array and an excellent radiator may each perform well alone yet create a difficult combined structure.

Heat capacity buys time, not permanent cooling

A thick piece of material can absorb heat while warming up. That may protect a system during a brief burst of computation. It does not dispose of the heat. Before the next burst, the stored energy still has to leave or the temperature will keep rising.

Consider an imagined platform that allows ten minutes of unusually heavy processing by using its thermal mass as a buffer. If it then needs forty minutes at reduced load to recover, its sustainable output is very different from the ten-minute benchmark.

A thermal buffer can be useful when work arrives in bursts or the environment changes predictably. It becomes misleading only when temporary performance is advertised as continuous capacity.

The same principle applies to batteries on the electrical side. Both storage and thermal mass can shift a problem in time. Neither automatically removes the need to balance the system over a complete operating cycle.

For a customer, the important number is how much work finishes through repeated cycles, including recovery. A business cannot sell the same temporary reserve forever without allowing it to recharge or cool.

Cooling is also a service decision

Suppose two hypothetical designs use the same processor. One carries a larger radiator and offers steadier performance. The other uses a smaller thermal system and sells interruptible jobs at a lower price.

Neither is inherently the better machine. Their value depends on the customers and complete costs. The first must earn enough from reliability to justify additional mass. The second must find work that remains useful despite pauses.

A third design might accept a lower processor power limit all the time, trading peak speed for a simpler thermal system. Depending on the workload, that could produce more useful lifetime output per launched kilogram than a design built around maximum chip performance.

These are examples of the design space, not predicted winners. They show how the cooling system helps define the product itself. The radiator does not merely support a promise made by the sales department; it helps determine which promises can honestly be made.

When the cooling design and customer contract disagree, physics will eventually enforce its own terms.

What would prove this?

Test a representative compute load in a thermal-vacuum environment, then in flight. Report radiator and component temperatures, rejected heat, incoming environmental loads, power spent on heat transport, and how often throttling occurs.

The strongest evidence is sustained operation through expected extremes—not a single cool photograph or a processor briefly reaching its maximum speed.

An orbital data center cannot escape thermodynamics. It can only use good engineering to make thermodynamics part of a viable business. The radiators may look less exciting than the chips, but they help decide whether those chips get to keep working.

Previous article · Next article